The purpose of this study is to identify predictive factors associated with missed diagnosis of B. pertussis–B. holmesii co-infection by assessing the analytical performance of a commercially available multiplexed PCR assay and by building a prediction model based on clinical signs and symptoms for detecting co-infections. This is a retrospective study on the electronic health records of all clinical samples that tested positive to either B. pertussis or B. holmesii from January 2015 to January 2018 at Geneva University Hospitals. Multivariate logistic regression was used to build a model for co-infection prediction based on the electronic health record chart review. Limit of detection was determined for all targets of the commercial multiplexed PCR assay used on respiratory samples. A regression model, developed from clinical symptoms and signs, predicted B. pertussis and B. holmesii co-infection with an accuracy of 82.9% (95% CI 67.9–92.8%, p value = .012), for respiratory samples positive with any of the two tested Bordetella species. We found that the LOD of the PCR reaction targeting ptxS1 is higher than that reported by the manufacturer by a factor 10. The current testing strategy misses B. pertussis and B. holmesii co-infections by reporting only B. holmesii infections. Thus, we advocate to perform serological testing for detecting a response against pertussis toxin whenever a sample is found positive for B. holmesii. These findings are important, both from a clinical and epidemiological point of view, as the former impacts the choice of antimicrobial drugs and the latter biases surveillance data, by underestimating B. pertussis infections during co-infections.